L4: 24 GB, what it runs and what it costs
Specs checked · Rental prices from 40m ago · Aliteq Team · How we compute this
The short answer
The NVIDIA L4 has 24 GB of VRAM and 300 GB/s of memory bandwidth (2023). At Q4_K_M quality or better with an 8K context it runs 31 of the open models we track; the largest that fits with room to spare is Qwen3.6 35B-A3B (MoE). The cheapest rental right now is $0.045/hr (Vast.ai, interruptible).
- VRAM
- 24 GB
- Bandwidth
- 300 GB/s
- Board power
- 72 W
- Launched
- 2023
Source: NVIDIA specification · bandwidth not yet confirmed by a second source, so we don't estimate speed from it
Buy it or rent it
| Provider | Type | From, per hour | Rent |
|---|---|---|---|
| Vast.ai | Spot (can be interrupted) | $0.045/hr | Vast.ai |
| Vast.ai | On demand | $0.104/hr | Vast.ai |
| Runpod | On demand | $0.590/hr | Runpod |
Lowest price per provider in our hourly snapshot (40m ago). Rent buttons are referral links. Every card side by side: cloud GPU prices.
Which AI models run on the L4
Computed by our VRAM engine for an 8K context: the best quality each model fits at in 24 GB. Speeds are upper-bound estimates from memory bandwidth, not benchmarks.
| Model | Best fit | VRAM used | ~tok/s |
|---|---|---|---|
| Qwen3.6 35B-A3B (MoE) | Q4_K_M | 21 GB88% | — |
| Ornith-1.5 35B-A3B (MoE) | Q4_K_M | 21 GB88% | — |
| Qwen3.5 35B-A3B (MoE) | Q4_K_M | 21 GB88% | — |
| DeepSeek R1 Distill 32B | Q5_0 | 24 GB100% · tight | — |
| Qwen2.5 Coder 32B | Q5_0 | 24 GB100% · tight | — |
| Qwen2.5 32B Instruct | Q5_0 | 24 GB100% · tight | — |
| Qwen3 32B | Q5_0 | 24 GB100% · tight | — |
| NVIDIA Nemotron 3 Nano 30B-A3B (MoE) | Q5_K_M | 22 GB90% · tight | — |
| Gemma 4 31B | Q5_K_M | 22 GB92% · tight | — |
| Qwen3 30B-A3B (MoE) | Q5_K_M | 22 GB91% · tight | — |
| Qwen3-Coder 30B-A3B (MoE) | Q5_K_M | 22 GB91% · tight | — |
| Qwen3.5 27B | Q6_K | 22 GB94% · tight | — |
Fits only below Q4 quality: Mixtral 8x7B (MoE) (Q3_K_M).
Too big for 24 GB (rent a bigger card instead): what it costs to run DeepSeek R1 Distill 70B, what it costs to run Qwen3-Coder-Next 80B-A3B (MoE), what it costs to run Qwen3-Next 80B-A3B (MoE), what it costs to run GLM-4.5-Air (MoE).
Check a model and context of your own in the VRAM calculator.
AI video on the L4
Alibaba's Wan 2.2 is an open video model you can run yourself in ComfyUI. With 24 GB the L4 fits:
- Wan 2.2 14B at 640x640 with fp8 weights (24 GB and up, per Runpod's Wan 2.2 guide)
- Wan 2.2 5B, with ComfyUI's native offloading (8 GB and up, per ComfyUI's Wan 2.2 docs)
What each setup needs, the cheapest card to rent and what one clip costs: cloud GPUs for ComfyUI with Flux and Wan 2.2. Against an app or an API: AI video cost per clip, app vs API vs rented GPU. Rather not run a model yourself? Hosted video models such as Seedance, Veo and Hailuo are compared in Kling AI alternatives, priced per clip.
Cards to compare it with
Every card by VRAM tier: all GPUs for gaming and local AI.